WASHH algorithm achieves perfect rank with whale-guided hybrid search

Learning-assisted algorithm design faces the challenge of making reliable search decisions under tiny evaluation budgets. A single metaheuristic often fails to adapt across diverse problems. To address this, researchers from multiple institutions propose WASHH (Whale-guided Adaptive Selection Hyper-Heuristic), which treats multiple search behaviors as selectable actions: WOA for exploitation, PSO-style memory, GWO-style leader averaging, DE-style variation, local coordinate search, and a novel anchor-guided refinement that exploits inexpensive reference configurations (e.g., box centers or default model settings) without bypassing black-box evaluation. An online reward controller dynamically allocates evaluations to the most promising behaviors based on observed improvements, making the algorithm self-adaptive.

WASHH was tested on 10 standard 30-dimensional benchmark functions (including Rastrigin, Griewank, Ackley, Sphere, and Rosenbrock) with 10 independent runs and 12,000 evaluations each. It achieved a perfect average rank of 1.10, meaning it was either the best or tied for best on every single function. Compared to standalone WOA, WASHH strictly improved on 8 functions and tied at the numerical optimum on Rastrigin and Griewank. This consistent superiority demonstrates that hybridizing multiple heuristics with an intelligent controller yields significantly more robust optimization than any single method.

The researchers further validated WASHH on a practical machine learning task: hyperparameter configuration of Support Vector Classifiers (SVC) for breast cancer diagnosis, under an extremely tight budget of 300 evaluations. WASHH obtained the lowest mean validation log loss among all compared optimizers (including standard WOA, PSO, and others), suggesting that anchor-aware selection hyper-heuristics are a practical, lightweight direction for algorithm configuration systems. This work opens the door for deploying sophisticated optimization in resource-constrained environments like medical AI or embedded systems.